Elegant Graphics for Data Analysis with ggplot2

Yann Abraham

BaselR 28.04.2010

Who is Yann Abraham

• Biochemist by training

• Bioinformatician by trade

• Pharma/Biotech

– Cellzome AG

– Novartis Pharma AG

• http://ch.linkedin.com/in/yannabraham

How to Represent Data

“A Picture is Worth a Thousand Words”

• Visualization is a critical component of data analysis

• Graphics are the most efficient way to digest large volumes of data & identify trends

• Graphical design is a mixture of mathematical and perceptual science

A Straightforward Way to Create Visualizations

• Grammar of Graphics provides a framework to streamline the description and creation of graphics

• For a given dataset to be displayed:– Map variables to aesthetics

– Define Layers• A representation (a ‘geom’) ie line, boxplot, histogram,…

• Associated statistical transformation ie counts, model,…

– Define Scales• Color, Shape, axes,…

– Define Coordinates

– Define Facets

Why Use ggplot2?

• Simple yet powerful syntax

• Provides a framework for creating any type of graphics

• Implements basic graphical design rules by default

An Example

• 4 cell lines where treated with a compound active against a class of enzymes

• Proteins where extracted and quantified using mass spectrometry

• Is there anything interesting?!?

Excel…

R…

R…

R…

R… (this could go on for hours)

…ggplot2!

ggplot2!

ggplot2!

ggplot2!

ggplot2!

qplot(Experiment_1,Experiment_2,data=comp,color=CELLLINE,facets=ISTARGET~CELLLINE)+coord_equal()+scale_x_log10()+scale_y_log10()

When Visualization Alone is Not Enough

• Some datasets are large multidimensional data structures

• Representing data from such structure requires data transformation

• R is good at handling large sets

• R functions for handling multidimensional sets are complex to use

Easy Data Transformation With plyr

• plyr provides wrappers around typical R operations

– Split

– Apply

– Combine

• plyr functions are similar to the by() function

Why use plyr?

• Simple syntax

• Predictable output

• Tightly integrated into ggplot2

• This comes at a price – somewhat slower than apply

An Example

• Given a set of raw data from a High Throughput Screen, compute the plate-normalized effect

The standard R way

plate.mean <- aggregate(hts.data$RAW,

list(hts.data$PLATE_ID),mean)

names(plate.mean) <- c(‘PLATE_ID’,’PLATE_MEAN’)

hts.data <- merge(hts.data,plate.mean)

hts.data$NORM<-hts.data$RAW/hts.data$PLATE_MEAN

The plyr way

hts.data <-ddply(hts.data,.(PLATE_ID),function(df) {

df$NORM<-df$RAW/mean(df$RAW)

return(df) }

)

Benefits of Using plyr & ggplot2

• Compact, straightforward syntax– Good basic output, complex options only required

for polishing

• Shifts focus from plotting to exploring– Presentation graphics can be created from there

at minimal cost

– Data transformation is intuitive

• Powerful statistics available– It’s R!

Some links…

• The Grammar of Graphics book by Leland Wilkinson

• The ggplot2 book by Hadley Wickham

– And the corresponding website

• A presentation about plyr by JD Long

– And his initial blog post

THANK YOU FOR YOUR ATTENTION!